CN111127409A - Train component detection method based on SIFT image registration and cosine similarity - Google Patents
Train component detection method based on SIFT image registration and cosine similarity Download PDFInfo
- Publication number
- CN111127409A CN111127409A CN201911283008.2A CN201911283008A CN111127409A CN 111127409 A CN111127409 A CN 111127409A CN 201911283008 A CN201911283008 A CN 201911283008A CN 111127409 A CN111127409 A CN 111127409A
- Authority
- CN
- China
- Prior art keywords
- image
- train
- cosine similarity
- detected
- sift
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/001—Industrial image inspection using an image reference approach
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30248—Vehicle exterior or interior
Abstract
The invention discloses a train component detection method based on SIFT image registration and cosine similarity, which comprises the following specific steps: 1. installing a linear array camera to shoot a train body and sending an image to a computer; 2. after reading the image, the computer intercepts the area of the component to be detected by using the relative coordinates of the image and calls a standard diagram of the corresponding component; 3. extracting SIFT features of the picture by using an SIFT algorithm; 4. performing feature matching by using a FLANN algorithm, and overlapping the image to be detected and the standard image on one image through affine transformation, thereby eliminating the image distortion of the image to be detected; 5. after registration, a mask is used for intercepting a superposed area, and a window is slid to detect defects; 6. and calculating the cosine similarity of the image of the area to be detected and the standard image and comparing the cosine similarity with each other to obtain a result. The invention has less hardware requirement and saves cost; the operation is simple and the use is convenient; and meanwhile, the speed and the accuracy of fault detection are improved.
Description
Technical Field
The invention belongs to the field of rail transit fault detection, and particularly relates to a train component detection method based on SIFT image registration and cosine similarity.
Background
The subway is a form of railway transportation, is an urban rail transit system mainly operated underground, and becomes an urban public transport means with the largest passenger flow in many cities in China. Although the production and operation systems of subway trains in China are very mature, a large number of exposed components on the bodies of the subway trains still have the risk of various faults. These unpredictable failures are time-to-time threatening the safety of the lives and property of millions of passengers and subway personnel each day. Therefore, the system capable of intelligently detecting the faults of the subway train is designed to have important significance for improving the safety of the subway system.
At present, the detection of subway train faults mostly depends on manual visual identification, and due to the fact that the number of parts is large and the picture contrast is limited, a large amount of labor is occupied in the work, and the train faults cannot be detected efficiently and accurately. Meanwhile, due to the high labor intensity of manual operation, the missed measurement or the mismeasurement is easily caused after the human body is fatigued. The intelligent special detection designed for the height valve, the sand sprinkling pipe and other parts on the subway train body can help to improve the safety of subway operation and liberate labor force. A simple and integrated intelligent detection program can also conveniently guide workers to process detected faults.
The subway train fault intelligent detection system has the advantages that subway train faults are intelligently detected, and the subway train safety guarantee system can be perfected and subway work groups are simplified by recognizing faults such as loosening and falling of core components such as altitude valves and sand sprinkling pipes, and the important significance is realized on development of subways.
The first prior art is as follows:
the invention discloses a train fastener looseness detection system based on an artificial identification image, and relates to the technical field of train fault detection. The system consists of a train speed measuring module, an imaging control module, an imaging module, an illuminating module and a data acquisition and processing module. The train speed measuring module measures the running speed of a train in real time, the imaging control module generates an imaging control signal according to the running speed of the train, the imaging module images the surface of the train according to the control signal, artificial marks are arranged on the surface of a train fastener, the imaging module shoots images and then inputs the images into the image acquisition and processing module, on the image acquisition and processing module, the shot artificial mark images are processed, whether the shape of the artificial mark on the surface of the fastener changes is judged through an image processing algorithm, namely, a linear mark is extracted and recognized, the included angle of the linear mark after dislocation is calculated, and whether the fastener looses is judged through the included angle.
Fig. 1 is a schematic view of embodiment 1 of the present invention, and fig. 2 is an example of a line at the top of a fastener. In the figure, 1-train, 2-track, 3-train speed measuring module, 4-imaging control module, 5-lighting module, 6-imaging module, 7-data acquisition and processing module, 8-linear mark, 9-screw rod and 10-nut.
Disadvantages of the first prior art
1. The invention needs a large number of additional hardware modules such as a train speed measuring module, an imaging control module and the like, which increases the cost of a large number of arrangements;
2. the method judges whether the fastener is loosened by identifying whether the shape of the artificial mark changes, and the shape of the artificial mark may change due to external reasons such as illumination, dirt and the like, so that false alarm occurs.
3. The invention adjusts the shooting frequency of the linear array camera through the speed measuring module to avoid stretching or compressing the image. However, the accuracy and real-time performance of the speed measurement module are extremely high, and the good effect is difficult to achieve at the same time of extremely high cost.
The second prior art is:
fig. 3 shows an image processing method, an image processing device, and a train fault detection system according to the invention. The image processing method comprises the following steps: acquiring a plurality of images of each vehicle in the train; and sequentially taking each image as a current processing image to perform abnormity detection processing. The abnormality detection processing is to sequentially perform feature comparison and key fault identification on the current image, or sequentially perform key fault identification processing and feature comparison processing, and when the feature comparison processing result and/or the key fault identification processing result indicate that the image is abnormal, determine that the current image is abnormal and mark an abnormal area on the current image; and providing the abnormal image for identifying the abnormal area to the vehicle inspection station equipment.
Wherein the feature comparison processing comprises extracting key features of the current image; comparing the key features of the current image with a pre-established reference feature model by taking the vehicle number as an index, calculating a matching rate, determining that the current image is abnormal when the matching rate is less than a preset threshold, and identifying an abnormal area on the current image; the key fault identification processing comprises the following steps: identifying train components in the current image; and calling a recognition algorithm corresponding to the current train component from a preset algorithm library and executing, and when the recognition result indicates that the train component has a fault, determining that the current image has an abnormality and identifying an abnormal area on the current image.
The second prior art has the defects
1. The method only achieves the mode of only providing the abnormal image (or the image with abnormal suspicion) to the detector, is not really intelligent abnormal fault detection, and still needs the detector to manually identify whether the fault exists or what kind of fault exists after the abnormal image is sent to the detector, so that the limited part of workload is only reduced;
2. in the invention, the component identification is applied to an edge detection method, which needs to set a threshold value, and is greatly influenced by factors such as illumination and the like, so that the stability of the method is poor;
3. in the embodiment, a support vector machine applied to a brake shoe drill rod loss identification algorithm needs a large number of positive and negative samples for training, and in actual application, negative samples of train images, namely train fault images, are difficult to obtain in a large number, so that the final test result of the method is difficult to obtain to expect;
4. the algorithm of the invention needs to separately process the feature recognition and the train component recognition, not only occupies a large amount of computer memory, but also causes long calculation time due to repeated image processing.
In conclusion, the prior art mostly detects train faults through manual detection. However, the workload is complicated and huge, so that the workers are difficult to avoid fatigue and lacked, thereby causing the risk of missing the report. Meanwhile, since it is difficult for the train to maintain a constant speed during the traveling, the photographed image is inevitably stretched or shortened.
Disclosure of Invention
In order to solve the technical problems, the invention provides a train component detection method based on SIFT image registration and cosine similarity.
The invention discloses a train component detection method based on SIFT image registration and cosine similarity, which comprises the following steps of:
step 1: and installing a linear array camera at a specified place to shoot the train body, and sending the image to a computer.
Step 2: because the train does not keep constant speed in the running process, the size of the shot train component (such as a sand pipe) image to be detected is not necessarily consistent with that of the standard image, namely the ground component in the image without faults, so after the image is read by the computer, the area of the component to be detected is firstly intercepted by using the relative coordinates of the image, and the standard image of the corresponding component is called. At this point, the part to be inspected in the image to be inspected is extracted.
And step 3: and extracting the SIFT features of the picture by using a SIFT algorithm. The SIFT features are local features of the image, the rotation, scale scaling and brightness change of the image are kept unchanged, the visual angle change, affine transformation and noise are kept stable to a certain degree, and the SIFT features are very suitable for extracting the features of train parts in the application.
And 4, step 4: the method comprises the steps of performing feature matching by using a FLANN algorithm (the FLANN is a nearest neighbor rapid library, and the most appropriate algorithm can be selected according to data to process a data set), and overlapping an image to be detected and a standard image on one image through affine transformation, so that image distortion of the image to be detected caused by non-uniform running of a train is eliminated.
And 5: after registration, a mask is used for intercepting a superposed area, and a window is slid to detect defects. The sliding window is to slide the windows with different sizes and proportions (aspect ratio) on the whole picture in a certain step length, and then detect the defects of the areas corresponding to the windows, so that the detection of the whole picture can be realized.
Step 6: comparing the image of the region to be detected extracted from the sliding window with the standard image, calculating the cosine similarity of the image and the standard image, and setting a threshold value for the similarity of the image and the standard image; if the similarity of a plurality of continuous adjacent sliding window areas is smaller than the threshold value, the component is indicated to be in failure, and related staff are informed to process.
Furthermore, the sensor of the line camera only has one line of photosensitive elements, so that an image with high scanning frequency and high resolution can be obtained.
Further, in each sliding window, it is identified whether the content in the window is similar to the standard graph in step 5.
The beneficial technology of the invention is as follows:
1. the invention can detect the faults of all trains in real time and report whether the faults occur to the working personnel in time.
2. The invention can carry out special detection on all train parts such as sand spreading pipes, collector shoes and the like in an integrated manner, has simple operation and is convenient for workers without computer foundation to use.
3. The invention only needs to install the linear array camera and the calculation processing equipment at a specified place, has no other hardware requirements and saves the cost.
Drawings
Fig. 1 is a schematic diagram of an embodiment of a prior art.
FIG. 2 is an example of a line on the top of a fastener of the first prior art.
Fig. 3 is a flowchart of a second prior art.
FIG. 4 is a schematic diagram of an image captured according to the present invention.
FIG. 5 is a flow chart of the present invention.
Fig. 6 is a sand pipe image.
Fig. 7 is a height valve image.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
The train component detection method based on SIFT image registration and cosine similarity is shown in FIG. 5, and takes the detection of a sand scattering pipe as an example, and comprises the following steps:
1: and (3) installing a linear array camera at a specified place to shoot the train body (the installation schematic diagram is shown in figure 4), and sending the image to a computer.
Here, the user needs to provide a positive sample of the sandpipe, i.e. a standard diagram without faults, to the program. For comparison with the image to be detected. The image of the train sanding pipe is captured by a line camera installed beside the train running track as shown in fig. 6, and then the image is sent to a computer for processing.
2: since the train does not keep constant speed in the running process, the size of the shot image of the sand pipe is not necessarily consistent with that of the standard image, namely the image without faults, and therefore after the image is read by the computer, the area of the part to be detected is firstly intercepted by using the relative coordinates of the image, and the standard image corresponding to the sand pipe is called. At this time, the sand pipe to be detected in the image to be detected is extracted.
3: and extracting the SIFT features of the picture by using a SIFT algorithm. The SIFT features are local features of the image, the rotation, scale scaling and brightness change of the image are kept unchanged, the visual angle change, affine transformation and noise are kept stable to a certain degree, and the SIFT features are very suitable for extracting the features of train parts in the application.
4: and overlapping the sand pipe to be detected and the standard image on a map by using a FLANN feature matching algorithm (FLANN is a nearest neighbor rapid library which can select the most appropriate algorithm to process a data set according to data) and affine transformation, thereby eliminating the image distortion of the sand pipe image to be detected caused by the non-uniform running of the train.
5: after registration, a mask is used for intercepting a superposed area, and a window is slid to detect defects. The sliding window is to slide the windows with different sizes and proportions (aspect ratios) on the whole picture in a certain step length, and then to identify the defects of the areas corresponding to the windows, so as to realize the detection of the whole picture.
6: comparing the extracted image of the sliding window area with the standard image, calculating the cosine similarity of the extracted image and the standard image, and setting a threshold value for the cosine similarity; and if the cosine similarity of the adjacent sliding window areas is higher than a preset threshold value, judging that the sand scattering pipe in the image has no fault. If the similarity is smaller than the threshold value, the component is indicated to be in failure, and related staff are informed to carry out processing.
Example 2
Fault detection is performed on the altitude valve. Similar to example 1, the user needs to provide a standard diagram of the altitude valve (as shown in fig. 7), then take an image through the line array camera, call the corresponding altitude valve standard image, and then perform the same processing and calculation to detect the fault of the altitude valve.
In theory, any train component that provides a standard map can be used for fault detection using this patent.
While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted for elements thereof without departing from the scope of the invention.
Claims (3)
1. The train component detection method based on SIFT image registration and cosine similarity is characterized by comprising the following steps of:
step 1: installing a linear array camera to shoot a train body and sending an image to a computer;
step 2: after reading the image, the computer firstly intercepts the area of the component to be detected by using the relative coordinates of the image and calls a standard diagram of the corresponding component;
and step 3: extracting SIFT features of the picture by using an SIFT algorithm;
and 4, step 4: performing feature matching by using a FLANN algorithm, and overlapping the image to be detected and the standard image on one image through affine transformation, thereby eliminating image distortion of the image to be detected caused by non-uniform running of the train;
and 5: after registration, a mask is used for intercepting a superposed area, and a window is slid to detect defects;
step 6: comparing the extracted image of the region to be detected with the standard image, calculating the cosine similarity of the extracted image and the standard image, and setting a threshold value for the cosine similarity; if the similarity of a plurality of adjacent sliding window areas is smaller than the threshold value, the component is indicated to be in failure, and related staff are informed to carry out processing.
2. The SIFT image registration and cosine similarity-based train component detection method as claimed in claim 1, wherein the line camera has only one row of photosensitive elements in its sensor.
3. The train component detection method based on SIFT image registration and cosine similarity as claimed in claim 1, wherein in each sliding window in step 5, it is identified whether the content in the window is similar to a standard image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911283008.2A CN111127409A (en) | 2019-12-13 | 2019-12-13 | Train component detection method based on SIFT image registration and cosine similarity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911283008.2A CN111127409A (en) | 2019-12-13 | 2019-12-13 | Train component detection method based on SIFT image registration and cosine similarity |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111127409A true CN111127409A (en) | 2020-05-08 |
Family
ID=70498685
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911283008.2A Pending CN111127409A (en) | 2019-12-13 | 2019-12-13 | Train component detection method based on SIFT image registration and cosine similarity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111127409A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111579560A (en) * | 2020-05-28 | 2020-08-25 | 成都国铁电气设备有限公司 | Defect positioning device and method used in tunnel |
| CN112801110A (en) * | 2021-02-01 | 2021-05-14 | 中车青岛四方车辆研究所有限公司 | Target detection method and device for image distortion correction of linear array camera of rail train |
| CN112991347A (en) * | 2021-05-20 | 2021-06-18 | 西南交通大学 | Three-dimensional-based train bolt looseness detection method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103512762A (en) * | 2012-06-29 | 2014-01-15 | 北京华兴致远科技发展有限公司 | Image processing method and device and train fault detection system |
| CN106952257A (en) * | 2017-03-21 | 2017-07-14 | 南京大学 | A kind of curved surface label open defect detection method based on template matches and Similarity Measure |
| CN107688806A (en) * | 2017-08-21 | 2018-02-13 | 西北工业大学 | A kind of free scene Method for text detection based on affine transformation |
| US20180273066A1 (en) * | 2017-03-24 | 2018-09-27 | Canadian Pacific Railway Company | Condition based maintenance of railcar roller bearings using predictive wayside alerts based on acoustic bearing detector measurements |
| CN109409404A (en) * | 2018-09-13 | 2019-03-01 | 西南交通大学 | A kind of high iron catenary radix saposhnikoviae bracing wire fault detection method based on deep learning |
-
2019
- 2019-12-13 CN CN201911283008.2A patent/CN111127409A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103512762A (en) * | 2012-06-29 | 2014-01-15 | 北京华兴致远科技发展有限公司 | Image processing method and device and train fault detection system |
| CN106952257A (en) * | 2017-03-21 | 2017-07-14 | 南京大学 | A kind of curved surface label open defect detection method based on template matches and Similarity Measure |
| US20180273066A1 (en) * | 2017-03-24 | 2018-09-27 | Canadian Pacific Railway Company | Condition based maintenance of railcar roller bearings using predictive wayside alerts based on acoustic bearing detector measurements |
| CN107688806A (en) * | 2017-08-21 | 2018-02-13 | 西北工业大学 | A kind of free scene Method for text detection based on affine transformation |
| CN109409404A (en) * | 2018-09-13 | 2019-03-01 | 西南交通大学 | A kind of high iron catenary radix saposhnikoviae bracing wire fault detection method based on deep learning |
Non-Patent Citations (4)
| Title |
|---|
| 王永强: "基于线阵相机扫描的地铁车顶关键部件检测研究", 《机车车辆工艺》 * |
| 王金龙等: "基于SIFT图像特征提取与FLANN匹配算法的研究", 《计算机测量与控制》 * |
| 胡方尚等: "基于印刷缺陷检测的图像配准方法研究", 《光学技术》 * |
| 邵进达等: "改进SIFT 算法结合两级特征匹配的无人机图像匹配算法", 《计算机科学》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111579560A (en) * | 2020-05-28 | 2020-08-25 | 成都国铁电气设备有限公司 | Defect positioning device and method used in tunnel |
| CN112801110A (en) * | 2021-02-01 | 2021-05-14 | 中车青岛四方车辆研究所有限公司 | Target detection method and device for image distortion correction of linear array camera of rail train |
| CN112801110B (en) * | 2021-02-01 | 2022-11-01 | 中车青岛四方车辆研究所有限公司 | Target detection method and device for image distortion correction of linear array camera of rail train |
| CN112991347A (en) * | 2021-05-20 | 2021-06-18 | 西南交通大学 | Three-dimensional-based train bolt looseness detection method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111322985B (en) | Tunnel clearance analysis method, device and system based on laser point cloud | |
| CN111127409A (en) | Train component detection method based on SIFT image registration and cosine similarity | |
| CN101893580B (en) | Digital image based detection method of surface flaw of steel rail | |
| CN105158257B (en) | Slide plate measurement method and device | |
| Dubey et al. | Maximally stable extremal region marking-based railway track surface defect sensing | |
| CN109190473A (en) | The application of a kind of " machine vision understanding " in remote monitoriong of electric power | |
| CN106991668B (en) | Evaluation method for pictures shot by skynet camera | |
| CN110567680B (en) | Track fastener looseness detection method based on angle comparison | |
| EP2697738A1 (en) | Method and system of rail component detection using vision technology | |
| CN106778833A (en) | Small object loses the automatic identifying method of failure under a kind of complex background | |
| CN104318582A (en) | Detection method for bad state of rotating double-lug component pin of high-speed rail contact network on basis of image invariance target positioning | |
| CN108846331B (en) | Video identification method for judging whether screw fastener of motor train unit chassis falls off or not | |
| CN106372667A (en) | Method for detecting adverse state of inclined sleeve part screws of high-speed train overhead line system | |
| CN111754460A (en) | Method, system and storage medium for automatically detecting gap of point switch | |
| Lu et al. | Automatic fault detection of multiple targets in railway maintenance based on time-scale normalization | |
| CN111881970A (en) | Intelligent outer broken image identification method based on deep learning | |
| CN106340019A (en) | Method for detecting adverse state of high-speed rail overhead line system inclined cable fixing hook | |
| CN113112501B (en) | Vehicle-mounted track inspection device and method based on deep learning | |
| CN109035249A (en) | A kind of parallel global threshold detection method of pipeline fault based on image procossing | |
| CN111724358A (en) | Concrete quality detection method and system based on image and convolutional neural network | |
| CN110728269B (en) | High-speed rail contact net support pole number plate identification method based on C2 detection data | |
| CN113591973B (en) | Intelligent comparison method for appearance state change of track plate | |
| CN105913008B (en) | Based on the assumption that the crowd's accident detection method examined | |
| Hashmi et al. | Computer-vision based visual inspection and crack detection of railroad tracks | |
| CN111553500B (en) | Railway traffic contact net inspection method based on attention mechanism full convolution network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200508 |